auto merge of #15113 : pnkfelix/rust/fsk-add-regression-test-for-ice-from-10846, r=alexcrichton

Includes a bit more comments than usual for a regression test; I felt like documenting Niko's diagnosis of the original problem here. 

Fix #15111 

r? anyone.

Author: bors

src/test/run-pass/regions-lifetime-nonfree-late-bound.rs

@@ -0,0 +1,40 @@
+// Copyright 2014 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// This is a regression test for the ICE from issue #10846.
+//
+// The original issue causing the ICE: the LUB-computations during
+// type inference were encountering late-bound lifetimes, and
+// asserting that such lifetimes should have already been subsituted
+// with a concrete lifetime.
+//
+// However, those encounters were occurring within the lexical scope
+// of the binding for the late-bound lifetime; that is, the late-bound
+// lifetimes were perfectly valid.  The core problem was that the type
+// folding code was over-zealously passing back all lifetimes when
+// doing region-folding, when really all clients of the region-folding
+// case only want to see FREE lifetime variables, not bound ones.
+
+pub fn main() {
+    fn explicit() {
+        fn test(_x: Option<|f: <'a> |g: &'a int||>) {}
+        test(Some(|_f: <'a> |g: &'a int|| {}));
+    }
+
+    // The code below is shorthand for the code above (and more likely
+    // to represent what one encounters in practice).
+    fn implicit() {
+        fn test(_x: Option<|f:      |g: &   int||>) {}
+        test(Some(|_f:      |g: &   int|| {}));
+    }
+
+    explicit();
+    implicit();
+}